Sued for Being Fired by AI? Meta Employees Reveal Harsh Reality: All Evidence Is in the Company's Hands

Reuters reported on Wednesday (22nd) that a group of former Meta (META-US) employees in the U.S. have filed a novel lawsuit, accusing the company of using discriminatory artificial intelligence (AI) tools to select employees for layoffs. However, the case also highlights the harsh reality workers face: even if they suspect they were fired by AI, it is extremely difficult to prove how companies actually use such technology.

Legal experts note that while it was previously expected that widespread corporate adoption of AI would trigger a wave of employment lawsuits, this anticipated surge has yet to materialize. One reason is that employees are typically unaware of how AI is applied in personnel decisions, and many have already signed arbitration agreements, waiving their right to sue in court.

U.S. Federal District Court Judge William Orrick last week rejected a request to block Meta from completing the termination process for 26 plaintiffs. In his ruling, he highlighted a fundamental barrier in AI discrimination cases: employees are 'not in the room where decisions are made,' making it difficult to quickly obtain the evidence needed to prove corporate wrongdoing.

The employees claim they were treated unfairly during the layoff process due to disabilities, medical leave, or family care leave. However, without knowing what data and tools Meta internally used, the plaintiffs currently struggle to refute the company's claims.

Arbitration Agreements Conceal AI Discrimination

Most U.S. workers have signed arbitration agreements with their employers, and the Meta plaintiffs are no exception. This means they cannot file class-action lawsuits, request jury trials, or seek large settlements through public court proceedings.

Companies typically argue that arbitration is faster and less costly than litigation; labor rights advocates, however, believe arbitration often favors employers and may deter employees from filing complaints. Because arbitration proceedings are not public, even if individual cases uncover that AI systems may produce discriminatory outcomes, the evidence is unlikely to be shared with other employees.

One of the few cases to reach court is a lawsuit against human resources software company Workday (WDAY-US), accused of using its software to illegally filter out job applicants at other companies based on race, age, and disability status. Since Workday does not sign arbitration agreements with job applicants at client companies, the case is not subject to such restrictions; Workday denies the allegations.

The agreement signed by Meta employees includes a common but narrowly defined exception allowing parties to petition the court for temporary relief to prevent irreversible actions by the other side. However, this exception is typically used in cases involving theft of trade secrets or poaching of clients or employees, and is rarely applied to the termination of at-will employees.

Employees Alleged AI Monitored Performance; Meta Denies

Orrick has dismissed the plaintiffs’ request for a temporary restraining order to halt layoffs but must still decide whether to issue a longer-lasting preliminary injunction to reinstate employees pending resolution of individual arbitration cases. A hearing is scheduled for August 24, and the losing party may appeal.

Orrick stated that if plaintiffs can present evidence showing whether and how AI was improperly used, he may reconsider his decision. Plaintiffs’ lawyers have acknowledged the difficulty of gathering evidence and are urging current and former employees who understand how Meta uses AI to select layoff targets to come forward, emphasizing that nearly all relevant information is held by Meta.

The plaintiffs allege that when deciding which positions to eliminate, Meta relied on tools tracking productivity and AI tool usage, placing employees who were absent due to health issues or family care at a disadvantage.

The complaint states that Meta uses multiple AI-assisted systems, including the large language model assistant 'Metamate,' the 'Second Brain' trained on employee data to track communications and documents, and productivity scores generated by scanning keyboard inputs, screen content, emails, and browsing history.

Meta, however, stated it announced the layoff of nearly 8,000 people earlier this year, and all such decisions were made by humans. It denies using AI tool usage metrics to select layoff candidates or conduct performance evaluations. Orrick noted that since plaintiffs currently cannot present evidence to refute this, he must temporarily accept Meta’s claims.